Re: patch -- (ABI-1-0-0-STABLE) XFT support

From: Joaquín Cuenca Abela (cuenca@pacaterie.u-psud.fr)
Date: Sun May 26 2002 - 15:29:37 EDT

  • Next message: Paul Rohr: "Re: patch -- (ABI-1-0-0-STABLE) XFT support"

    On Sun, 2002-05-26 at 21:09, Paul Rohr wrote:
    > At 11:57 PM 5/20/02 +0200, Joaquín Cuenca Abela wrote:
    > >I've added XFT support for AbiWord.
    > >
    > > [...]
    > >
    > >New problems:
    > >
    > >1) It's slower.
    >
    > Oh dear. It'd be wonderful to get anti-aliased text by default on Unix too,
    > but drawing speed is *critical* to the look and feel of a word processor. I
    > hope that this speed problem is a temporary artifact of your current
    > implementation, and not an inherent weakness in recent XFT code.

    yes, the problem with the speed has disappeared when I fixed the crash
    (the 3rd problem). As Martin pointed out, I was opening a XftFont when
    I was asked, and I was asked to open fonts at normal size, and with a 15
    times bigger size (to do the calculations in layout units).

    I've done a temporal hack (more or less following the hack present in
    the gdk code) until we solve all the WYSIWYG problems we have, and now
    speed is more than ok.

    The display speed is the same one than with the gdk backend, but the
    calculation for the size of the glyphs is taking a bit more of time than
    in gtk, but it's not an inherent problem with Xft, I'm just not caching
    the glyph indexes when I do the calculations in layout mode.

    I've not yet fixed it because

    1) the speed it's absolutely acceptable.
    2) I'm first trying to get correctness (the only thing that I'm lacking
    afaik is the insert symbol dialog box).

    This code will run slower in X servers without the render extension, but
    even there you will get something as 22000 glyphs/sec (10 times more
    with the render extension).

    > IIRC, we looked at using FreeType back in 1998, and at the time, drawing
    > performance was unacceptably poor. I had assumed that problem had gone away
    > after four years of optimization. Has it?

    It seems so. If you feel better, my girlfriend is absolutely not
    capable to find any difference in speed... it's only that I know where
    to look :) (ie, copy & paste of big chucks of text).

    Cheers,

    -- 
    Joaquín Cuenca Abela
    cuenca@pacaterie.u-psud.fr
    


    This archive was generated by hypermail 2.1.4 : Sun May 26 2002 - 15:29:58 EDT